Software Engineering Courses (SECourses)•3y ago

those models are nasty

Furkan Gözükara SECoursesOP•10/8/23, 1:22 PM

i think i would try sdxl

EElisoli Adetailer used here? I prefer resaults without adetailer as it keeps the origina...

Furkan Gözükara SECoursesOP•10/8/23, 1:22 PM

adetailer were used for realistic images

Furkan Gözükara SECoursesOP•10/8/23, 1:22 PM

for styled i didnt use

Elisoli•10/8/23, 1:22 PM

ohh and not used for stylized, very good idea

EElisoli ohh and not used for stylized, very good idea

Furkan Gözükara SECoursesOP•10/8/23, 1:23 PM

yes

Elisoli•10/8/23, 1:23 PM

also should we then use Text Encoder training or not? kept the suspence going on that twit

Elisoli•10/8/23, 1:23 PM

looks pretty identical

Furkan Gözükara SECoursesOP•10/8/23, 1:23 PM

this one is amazing

EElisoli also should we then use Text Encoder training or not? kept the suspence going o...

Furkan Gözükara SECoursesOP•10/8/23, 1:23 PM

i think looks like not very necessary

Furkan Gözükara SECoursesOP•10/8/23, 1:23 PM

but can be used too :d

rafraf•10/8/23, 1:24 PM

@Dr. Furkan Gözükara does NOT-raining text encoder save memory and speed? Is it worth turning off TE learning?

rafraf•10/8/23, 1:25 PM

Oh just noticed someone asked the same question

Rrafraf @Dr. Furkan Gözükara does NOT-raining text encoder save memory and speed? Is it ...

Furkan Gözükara SECoursesOP•10/8/23, 1:25 PM

it is turned off by default

Furkan Gözükara SECoursesOP•10/8/23, 1:25 PM

it increases vram slighty and reduces training speed slightly

rafraf•10/8/23, 1:26 PM

Also, where do you explain the block training --block_lr 1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5,1e-5

Elisoli•10/8/23, 1:26 PM

I had Very good success using control net Depth, Canny and openpose to upscale and bring out details in an old photo from this

Rrafraf Also, where do you explain the block training --block_lr 1e-5,1e-5,1e-5,1e-5,1e...

Furkan Gözükara SECoursesOP•10/8/23, 1:26 PM

use config from patreon

FFurkan Gözükara SECourses it increases vram slighty and reduces training speed slightly

rafraf•10/8/23, 1:26 PM

Ok good to know

EElisoli I had Very good success using control net Depth, Canny and openpose to upscale a...

Furkan Gözükara SECoursesOP•10/8/23, 1:26 PM

nice

Elisoli•10/8/23, 1:26 PM

I wish there was a better Canny model for detecting edges

Elisoli•10/8/23, 1:26 PM

had to mess with it a lot

EElisoli I wish there was a better Canny model for detecting edges

Furkan Gözükara SECoursesOP•10/8/23, 1:27 PM

ye we need better sdxl controlnet

Elisoli•10/8/23, 1:27 PM

also waiting for QR control net for sdxl, I cant seem to properly upscale 1.5 images from 512 to higher ones without losing details

FFurkan Gözükara SECourses those models are nasty

Stefanoid9343•10/8/23, 1:34 PM

Thank you so much!

Stefanoid9343•10/8/23, 1:39 PM

@Dr. Furkan Gözükara this is the new one that replace the real vision one? https://youtu.be/AY6DMBCIZ3A?feature=shared can I do it with runpod?

YouTubeSECourses

First Ever SDXL Training With Kohya LoRA - Stable Diffusion XL Trai...

Updated for SDXL 1.0. How to install #Kohya SS GUI trainer and do #LoRA training with Stable Diffusion XL (#SDXL) this is the video you are looking for. I have shown how to install Kohya from scratch. The best parameters to do LoRA training with SDXL. How to use Kohya SDXL LoRAs with ComfyUI. How to do checkpoint comparison with SDXL LoRAs and m...